One of the most important discoveries from extragalactic observations at mid- and far-infrared wavelengths has been the identification of a class of "Luminous Infrared Galaxies" (LIGs), objects that emit more energy in the infrared ( ~ 5-500µm) than at all other wavelengths combined (see  for a comprhensive review). The first all-sky survey at far-infrared wavelengths carried out in 1983 by the Infrared Astronomical Satellite (IRAS) resulted in the detection of tens of thousands of galaxies, the vast majority of which were too faint to have been included in previous optical catalogs. It is now clear that part of the reason for the large number of detections is the fact that the majority of the most luminous galaxies in the Universe are extremely dusty. Previous assumptions, based primarily on optical observations, about the relative distributions of different types of luminous galaxies - e.g. starbursts, Seyferts, and quasi-stellar objects (QSOs) - need to be revised.
Galaxies bolometrically more luminous than ~ 4L* (i.e. Lbol 1011 L) appear to be heavily obscured by dust. Although luminous infrared galaxies (hearafter LIGs: Lir > 1011 L) are relatively rare objects, reasonable assumptions about the lifetime of the infrared phase suggest that a substantial fraction of all galaxies with LB > 1010 L pass through such a stage of intense infrared emission .
A comparison of the luminosity function of infrared bright galaxies with other classes of extragalactic objects in the local universe is shown in Figure 1. At luminosities below 1011 L, IRAS observations confirm that the majority of optically selected objects are relatively weak far-infrared emitters. Surveys of Markarian galaxies confirm that both Markarian starbursts and Seyferts have properties (e.g. f60 / f100 and Lir / LB ratios) closer to infrared selected samples as does the subclass of optically selected interacting galaxies. However because the most luminous galaxies are enshrouded in dust, relatively few objects in optically selected samples are found with Lir > 1011.5 L.
Figure 1. Galaxy luminosity function of Infrared Galaxies compared with other extragalactic objects in the local universe. Among the most luminous galaxies (Lbol > 1011.5 L), infrared galaxies selected from the IRAS survey outnumber optically selected Seyferts and quasars. For references see .
The high luminosity tail of the infrared galaxy luminosity function is clearly in excess of what is expected from the Schechter function. For Lbol = 1011 - 1012 L, LIGs are as numerous as Markarian Seyferts and ~ 3 times more numerous than Markarian starbursts. Ultraluminous infrared galaxies (hereafter ULIGs: Lir > 1012 L) appear to be ~ 2 times more numerous than optically selected QSOs, the only other previously known population of objects with comparable bolometric luminosities.
Although LIGs comprise the dominant population of extragalactic objects at Lbol > 1011 L, they are still relatively rare. For example, Figure 1 suggests that only one object with Lir > 1012 L will be found out to a redshift of ~ 0.033, and indeed, Arp220 (z = 0.018) is the only ULIG within this volume. The total infrared luminosity from LIGs in the IRAS Bright Galaxy Survey (BGS) is only ~ 6% of the infrared emission in the local Universe .
There are preliminary indications that ULIGs have been more numerous in the past. Comparison of the space density of nearby ULIGs with the more distant population provides evidence for possible strong evolution in the luminosity function at the highest infrared luminosities. Assuming pure density evolution of the form (z) (1 + z)n,  found n ~ 7 ± 3 for a complete flux-limited sample of ULIGs.
The infrared properties for the complete IRAS Bright Galaxy Sample have been summarized and combined with optical data to determine the relative luminosity output from galaxies in the local Universe at wavelengths ~ 0.1-1000µm . Figure 2 illustrates how the shape of the mean spectral energy distribution (SED) varies for galaxies with increasing total infrared luminosity. Systematic variations are observed in the mean infrared colors; the ratio f60 / f100 increases while f12 / f25 decreases with increasing infrared luminosity. Figure 2 also illustrates that the observed range of over 3 orders of magnitude in Lir for infrared-selected galaxies is accompanied by less than a factor of 3-4 change in the optical luminosity.
Figure 2. Variation of the mean Spectral Energy Distribution (from submillimeter to UV wavelengths) with increasing Lir for a 60µm sample of infrared galaxies. ( Insert) Examples of the subset ( ~ 15%) of ULIGs with "warm" infrared color (f25 / f60 > 0.3). Three objects (1 - the powerful Wolf-Rayet galaxy IRAS01002-2238, 2 - the "infrared QSO" IRAS07598+6508, 3 - the optically selected QSO IZw1) are shown in the inset. For references see .
 showed that a small but significant fraction of ULIGs, those with "warm" (f25 / f60 >0.3) infrared colors, have SEDs with mid-infrared emission ( ~ 5-40µm) over an order of magnitude stronger than the larger fraction of "cooler" ULIGs. These warm galaxies (Figure 2 insert), which appear to span a wide range of classes of extragalactic objects including powerful radio galaxies (PRGs: L408MHz 1025 W Hz-1) and optically selected QSOs, have been used as evidence for an evolutionary connection between ULIGs and QSOs (e.g. [19, 20]).
There is a strong correlation between the broad band colors (from optical to far-infrared) and morphological type . In particular, the fraction of objects that are interacting/merger systems appears to increase systematically with increasing infrared luminosity. The imaging surveys of objects in the local universe [19, 12] have shown that the fraction of strongly interacting/merger systems increases from ~ 10% at log (Lir / L) = 10.5-11 to ~ 100% at log (Lir / L) > 12. In pannel (c) of Figure 3 is shown the "Super-antennae", which is the prototype of ULIG . ISO observations  have shown that more than 98% of the mid-infrared flux from this object comes from the southern component which hosts a Seyfert 2 nucleus.
Figure 3. Well-studied mergers: (a) NGC4038/39 (Arp244 = "The Antennae"); (b) NGC7252 (Arp226 = "Atoms for Peace"); (c) IRAS19254-7245 ("The Super Antennae"); (d) IC4553/54 (Arp220). The two at the top are LIGs whereas the two at the bottom are ULIGs. Contours of HI 21-cm line column density (black) are superimposed on deep optical (r-band) images. Inserts show a more detailed view in the K-band (2.2µm) of the nuclear regions of NGC4038/39, NGC7252, and IRAS19254-7245, and in the r-band (0.65µm) of Arp220. White contours represent the CO(1 0) line integrated intensity as measured by the OVRO millimeter-wave interferometer. No HI or CO interferometer data are available for the southern hemisphere object IRAS19254-7245. The scale bar represents 10kpc.
From the detailed studies of nearby ultraluminous infrared galaxies the following conclusions were reached. 1) They are mergers of evolved gas-rich giant spiral galaxies (e.g. Milky Way with Andromeda), and not "primival" galaxies. 2) To boost the luminosity above 1012 L the nuclei must have approached at least 10 kpc, namely, they are advanced mergers. 3) Due to the gravitational impact the interstellar gas decouples from the stars and large amounts of interstellar matter fall at high rates to the central region. This is the condition to produce a nuclear starburst, and/or feed a supermassive black hole at super-Eddington accretion rates. To produce such large accretion rates, the gravitational potential wheels of massive buldges are needed.
A workshop on the question concerning the ultimate source of energy (starbursts versus AGN's) took place in Ringberg on October 1998. Below 2 1012 L starbursts dominate the energy budget, but above 3 1012 L AGN's seem to be always present and become an important source of energy. In this respect it is interesting to note that it is found with ISO that in the prototype Seyfert 2 galaxy NGC 1068, about 80% of the mid-infrared flux between 4 and 18 µm comes from the AGN .
A caveat for the subject of this conference is that the pre-encounter objects that merged at high redshifts must have been different from the metal-rich evolved galaxies merging at present. Another caveat is that ultraluminous IR galaxies at high redshifts may be very difficult to detect using the Lyman break technique. Due to the large amounts of dust in ultraluminous objects, very little or none continuum leaks out at ultraviolet wavelengths. Therefore, surveys with submillimeter arrays as ALMA will be needed to detect ultraluminous galaxies at high redshifts.